Pearson vs Fisher
Locked
 
|
Hi all,
I posted a question a few weeks ago regarding how to know when to use the Fisher's Exact test vs using the Exact Pearson Chi-Sq. I was wondering if anyone had any additional guidance or references on that issue, especially when dealing with larger than 2x2 tables. In addition, when computationally feasible, are exact tests always preferable to Monte Carlo simulations or are there situations when a MC simulation should be used? I have included a sample table below with the results from significance testing to show the issue with Exact Tests I described in the previous post. If anyone is able to quickly run this and tell me if they get similar/different results from my own I would appreciate that. My exact tests seem to be giving weird results. Thank you for any help you are able to provide. Best, Walker Table: Col 1 Col 2 Row 1: 24 15 Row 2: 3 1 Row 3: 1 2 Results: Exact Monte Carlo Pearson (2 sided) .377 .680 Fisher (2 sided) 1.000 .680 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
Administrator
|
Ian Campbell argues (in a Statistics in Medicine article) that for the 2x2 table, except when both sets of marginal totals are fixed, the N-1 chi-square performs better than the Fisher-Irwin test (his name for the Fisher exact test, because Irwin also developed it, independently of Fisher). There is lots of good info on Campbell's website, including an online calculator.
http://www.iancampbell.co.uk/twobytwo/twobytwo.htm There may be extensions of the Fisher-Irwin test for tables larger than 2x2, but they would not be called the "Fisher exact test", I don't think. That name is reserved for the usual test on 2x2 tables. Re your other question, I suspect there are situations where "exact" is not best. This argument can certainly be made for things like confidence intervals on proportions. E.g., http://www.iancampbell.co.uk/twobytwo/twobytwo.htm HTH.
--
Bruce Weaver bweaver@lakeheadu.ca http://sites.google.com/a/lakeheadu.ca/bweaver/ "When all else fails, RTFM." PLEASE NOTE THE FOLLOWING: 1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above. 2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/). |
|
|
In reply to this post by Walker
What release of SPSS are you running?
Is anyone from SPSS Support reading? I believe that SPSS has a bug in its calculation and reporting of the exact values. The Monte Carlo p-values are around .680. Running in Cytel's StatXact, the exact p-values are essentially that -- 0.6789. Does anyone have any other non-SPSS software that reports exact p-values for the r x c table? Tony Babinec [hidden email] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Walker Sent: Friday, July 08, 2011 9:14 AM To: [hidden email] Subject: Pearson vs Fisher Hi all, I posted a question a few weeks ago regarding how to know when to use the Fisher's Exact test vs using the Exact Pearson Chi-Sq. I was wondering if anyone had any additional guidance or references on that issue, especially when dealing with larger than 2x2 tables. In addition, when computationally feasible, are exact tests always preferable to Monte Carlo simulations or are there situations when a MC simulation should be used? I have included a sample table below with the results from significance testing to show the issue with Exact Tests I described in the previous post. If anyone is able to quickly run this and tell me if they get similar/different results from my own I would appreciate that. My exact tests seem to be giving weird results. Thank you for any help you are able to provide. Best, Walker Table: Col 1 Col 2 Row 1: 24 15 Row 2: 3 1 Row 3: 1 2 Results: Exact Monte Carlo Pearson (2 sided) .377 .680 Fisher (2 sided) 1.000 .680 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
Thanks Bruce and Anthony. I will look into those resources. I am running Version 19.0.0.1 Windows 7 64-bit. After some investigation and talking with the folks at IBM we have discovered that incorrect exact test results are a known issue for 64-bit machines. They told us that they are hoping to resolve it by August. It is not, apparently, a problem on 32-bit machines. We tested two 32-bits here and the results from those matched each other and what Anthony got using StatXact. Thanks again.
Walker -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Anthony Babinec Sent: Friday, July 08, 2011 12:45 PM To: [hidden email] Subject: Re: Pearson vs Fisher What release of SPSS are you running? Is anyone from SPSS Support reading? I believe that SPSS has a bug in its calculation and reporting of the exact values. The Monte Carlo p-values are around .680. Running in Cytel's StatXact, the exact p-values are essentially that -- 0.6789. Does anyone have any other non-SPSS software that reports exact p-values for the r x c table? Tony Babinec [hidden email] -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Walker Sent: Friday, July 08, 2011 9:14 AM To: [hidden email] Subject: Pearson vs Fisher Hi all, I posted a question a few weeks ago regarding how to know when to use the Fisher's Exact test vs using the Exact Pearson Chi-Sq. I was wondering if anyone had any additional guidance or references on that issue, especially when dealing with larger than 2x2 tables. In addition, when computationally feasible, are exact tests always preferable to Monte Carlo simulations or are there situations when a MC simulation should be used? I have included a sample table below with the results from significance testing to show the issue with Exact Tests I described in the previous post. If anyone is able to quickly run this and tell me if they get similar/different results from my own I would appreciate that. My exact tests seem to be giving weird results. Thank you for any help you are able to provide. Best, Walker Table: Col 1 Col 2 Row 1: 24 15 Row 2: 3 1 Row 3: 1 2 Results: Exact Monte Carlo Pearson (2 sided) .377 .680 Fisher (2 sided) 1.000 .680 ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD ===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD |
|
|
|
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |